JP2005251091A - Data processor, data processing method, and data processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a data processor available for a data retrieving device for validly utilizing information related with the classification of data to be retrieved for data retrieval or the storage of the data to be retrieved. <P>SOLUTION: A keyword generating part 16 generates keywords from each data of a data group 111 by using a Japanese analytic engine 17, and outputs those keywords to a keyword classifying part 162. The keyword classifying part 162 decides classification items by clustering the keywords, and stores the keywords in a data store 11 based on the decided classification items. Clustering by the keyword classifying part 162 is operated based on the classified data group 111. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a data processing device, a data processing method, and a data processing program.

  2. Description of the Related Art Data search devices that perform searches on a large number of data using queries are widely used. In such a data search apparatus, if a query used for search is appropriate, a relatively small amount of data including desired data can be obtained as a search output. However, selection of an appropriate query requires prior knowledge of desired data, and needs to be familiar to some extent with a data search algorithm in the data search apparatus. For this reason, in a data search apparatus, a query including an inappropriate keyword is often used for data search, and a search result including a huge amount of noise is output, or desired data is not included in the search output. .

  In order to solve this problem, a technique is known in which knowledge about words is used for data search and storage of search target data.

  For example, Patent Literature 1 discloses a technique for presenting related words of an input word using a graph including knowledge about the word. Patent Document 2 discloses a technique for classifying and accumulating data using a region ontology storing knowledge about words. Further, Patent Document 3 discloses a technique for classifying and storing data using an ontology dictionary including knowledge about words.

JP 2000-32394 A JP 2000-276487 A JP 2001-229177 A

  However, in the techniques of Patent Documents 1 to 3, when the search target data is classified into a classification item group including a plurality of classification items, the information on the classification is used for data search and accumulation of search target data. I couldn't.

  The present invention has been made to solve this problem, and an object of the present invention is to provide a data processing apparatus that can effectively use information related to classification of data to be searched for data search and storage of data to be searched. To do.

  In order to solve the above-mentioned problem, the invention of claim 1 is a data processing device, and a generating means for generating a second data group from a first data group classified into a classification item group including a plurality of classification items; Classification means for classifying the second data group into the classification item group based on the classified first data group, and the classification to which new data not included in the first data group and the second data group belongs And determining means for determining a classification item in the item group based on the classified second data group.

  According to a second aspect of the present invention, in the data processing device according to the first aspect, each data of the first data group is text data, and each data of the second data group is text data of the first data group. It is characterized by being text data corresponding to an index.

  According to a third aspect of the present invention, in the data processing device according to the second aspect, the first data group is a data group related to a document, and the second data group is a data group related to an index of the document. Features.

  According to a fourth aspect of the present invention, in the data processing device according to the second aspect, the first data group is a data group related to an incident generated by a medical practice, and the second data group is data related to an index of the incident. It is a group.

  According to a fifth aspect of the present invention, in the data processing device according to any one of the first to fourth aspects, the second data group is partial data of the first data group.

  According to a sixth aspect of the present invention, in the data processing device according to the second aspect, the generating means includes a morpheme analyzing means for dividing a character string described in the text data into word groups, and the second from the word groups. And a construction means for constructing a data group.

  A seventh aspect of the present invention is the data processing apparatus according to the sixth aspect, wherein the construction unit includes a unit for removing duplicate words in the word group.

  The invention according to claim 8 is the data processing device according to claim 6 or claim 7, wherein the construction means specifies an unnecessary word in the word group according to a predetermined unnecessary word specifying rule, and the unnecessary word is determined. It has the means to remove, It is characterized by the above-mentioned.

  According to a ninth aspect of the present invention, in the data processing device according to any one of the sixth to eighth aspects, the classification unit calculates a weight of each data of the second data group in each data of the first data group. A calculation means for calculating according to a predetermined weight calculation rule is provided, and the weight is used for classification of the second data group.

  A tenth aspect of the present invention is the data processing apparatus according to the ninth aspect, wherein the weight is an importance level of data determined by a predetermined importance level determination rule.

  The invention according to claim 11 is the data processing apparatus according to claim 10, wherein the importance is a product of an appearance frequency of an index word and a reciprocal of the number of documents including the index word. .

  According to a twelfth aspect of the present invention, in the data processing device according to any one of the first to eleventh aspects, the classification means corresponds each data of the second data group to a classification item of the first data group. Vectorizing means for converting into a vector having a component to be used, and the vector is used for classification of the second data group.

  According to a thirteenth aspect of the present invention, in the data processing device according to the twelfth aspect of the present invention, the data of the second data group is based on a weight expressing the degree of relevance to each classification item of the first data group. Each component of the vector is determined.

  A fourteenth aspect of the present invention is the data processing device according to any one of the first to thirteenth aspects, wherein the classification means uses a neural network for classification.

  According to a fifteenth aspect of the present invention, in the data processing device according to the fourteenth aspect, the neural network is a self-organizing map.

  A sixteenth aspect of the present invention is the data processing device according to any one of the first to fifteenth aspects, wherein the new data is data added to the first data group.

  The invention according to claim 17 is the data processing device according to any one of claims 1 to 15, wherein the new data is generated from a query used for a search for the first data group. .

  According to an eighteenth aspect of the present invention, in the data processing device according to the seventeenth aspect of the present invention, the data processing device further includes an extracting unit that extracts data of the second data group belonging to the classification item to which the new data belongs.

  The invention according to claim 19 is the data processing device according to claim 18, further comprising output means for outputting auxiliary information relating to the new data, wherein the auxiliary information is included in another classification item to which the new data belongs. It is created based on the data of the second data group.

  According to a twentieth aspect of the present invention, in the data processing device according to the first aspect of the present invention, the data processing apparatus further includes an input unit that allows an operator to input the new data.

  The invention of claim 21 is a data processing method, the step of generating a second data group from a first data group classified into a classification item group including a plurality of classification items, and the classified first data group A classification process for classifying the second data group into the classification item group, and a classification item in the classification item group to which new data not included in the first data group and the second data group belongs. And a determination step of determining based on the classified second data group.

  The invention according to claim 22 is a data processing program, and when the program is executed by a computer, the computer generates a second data group from the first data group classified into a classification item group including a plurality of classification items. A generating step, a classification step of classifying the second data group into the classification item group based on the classified first data group, and a new process not included in the first data group and the second data group A determination step of determining a classification item in the classification item group to which the data belongs based on the classified second data group is performed.

  According to the inventions of claims 1 to 22, the second data group is classified based on the classified first data group, and the classification item to which the new data belongs is based on the classified second data group. Therefore, the information on the classification of the first data group can be effectively used for processing new data.

  According to the invention of claim 3, since the classification item of the document to which the new data belongs is determined, the classification item of the document can be used for the processing of the new data.

  According to the invention of claim 4, since the incident classification item to which the new data belongs is determined, the incident classification item can be used for processing the new data.

  According to the ninth to thirteenth aspects of the present invention, since each data of the second data group is digitized, the classification process of the second data group becomes easy.

  According to the twelfth or thirteenth aspect of the present invention, since each data of the second data group is vectorized, the classification process of the second data group is facilitated.

  According to the invention of claim 18 or claim 19, since the data of the other second data group included in the classification item to which the new data belongs is extracted, the extracted data can be used.

  According to the invention of claim 19, since the auxiliary information created based on the data of the other second data group included in the classification item to which the new data belongs is output, the operator can use the auxiliary information. Become.

  According to the twentieth aspect, it is possible to process new data input by the operator.

  The data search device 1 of this embodiment performs a search for the data group 111 stored in the data store 11 using the input query. As shown in FIG. 2, in the data search device 1, the data group 111 is stored in the data store 11 in a state of being classified into a plurality of classification items. The data store 11 of the data search apparatus 1 stores a keyword group 112 extracted from the data group 111 in a state of being classified into the same classification items as the data group 111.

  Furthermore, the data search device 1 that is also a data processing device determines the classification item to which the additional data belongs by using a keyword included in the additional data even when the classification item is not specified when registering additional data in the data group 111. Is possible. In addition, when a query is input, the data search device 1 can generate a hint that is auxiliary information related to the query using a keyword included in the query. The data search apparatus 1 can process both text data and binary data (image data, movie data, etc.). In the following, text data in which the data group 111 describes a Japanese character string is used. The description will be made assuming that

<Hardware configuration>
FIG. 1 is a block diagram showing a hardware configuration of a computer 4 for realizing the data search device 1 of the present embodiment.

  The computer 4 includes a CPU 42, a memory 43, a graphics adapter 44, an input device 45, and a storage device 46 connected via a bus 41 and the like. A display 47 serving as an output device that provides the user interface 12 is connected to the graphics adapter 44. The input device 45 that provides the user interface 12 includes a keyboard 451 and a pointing device 452.

  An operating system (hereinafter also referred to as “OS”) 461 that defines the basic operation of the computer 4 is installed in the storage device 46 configured by a hard disk drive or the like. A data search program 462 that is executed under the management of the OS 461 is also installed in the storage device 46. In the data search device 1, the computer 4 executes the data search program 462, thereby realizing the function of each functional block in FIGS. 2, 5-6, and 8-9 described later.

  In addition, although the example in which the data search device 1 is realized by the single computer 4 has been described above, it is not hindered that the data search device 1 is realized by a plurality of computers connected to the network.

<Functional configuration>
○ Overall composition;
FIG. 2 is a block diagram showing the overall functional configuration of the data search apparatus 1.

Data group 111 and keyword group 112 are stored in data store 11 of data search device 1 realized using storage device 46. The data group 111 is a set of data to be searched in the data search device 1. The keyword group 112 is a set of keywords of each data constituting the data group 111 and has a data system different from that of the data group 111. The keyword is an index or an index characterizing data, and is preferably partial data of data constituting the data group 111. As a result, keywords can be extracted from each data in the data group 111, so that keywords can be easily generated. The data group 111 and the keyword group 112 are stored in the data store 11 in a state of being classified into a classification item group including the classification items G ₀ , G ₁ ,..., G _Nc−1 . FIG. 3 shows an example of a state in which the data group 111 {D ₀ , D ₁ ,..., D _p } is classified into the classification items G ₀ , G _{1 ,.} Yes. Further, in FIG. 4, the keyword group _{112 {K 0, K 1,} ···, K Nk} classification item G _0, G _1, · · ·, are examples of conditions that are classified as G _Nc-1 shows Has been. The data model of the data group 111 and the keyword group 112 is not limited to the relational type, and may be a hierarchical type or a network type.

  The data search apparatus 1 includes a user interface 12 that acquires input data newly given from the outside such as an operator and presents output data to the outside such as the operator. The input data includes a query used for searching the data group 111 and additional data added to the data group 111. The output data includes a search result obtained by a search for the data group 111 and a hint related to the query. The query and additional data acquired by the user interface 12 are output to the query processing engine 13 and the additional data processing engine 14, respectively.

  The query processing engine 13 analyzes the given query, generates a keyword, and outputs it to the data search engine 15. In addition, the query processing engine 13 acquires another keyword (hereinafter also referred to as “same classification keyword”) belonging to the classification item to which the keyword generated by accessing the data store 11 belongs. Further, the query processing engine 13 generates the above hint based on the same classification keyword and outputs it to the user interface 12.

  The data search engine 15 accesses the data store 11 and acquires data including the given keyword. Further, the data search engine 15 outputs the acquired data to the user interface 12 as a search result.

  The additional data processing engine 14 analyzes the given additional data and generates a keyword. Further, the additional data processing engine 14 accesses the data store 11, specifies the classification item to which the generated keyword belongs, and determines the classification item to which the additional data belongs. The additional data for which the classification item is determined is stored in the data store 11 according to the classification item.

  The data processing engine 16 generates a keyword group 112 from the data group 111 and classifies the keyword group 112 based on the classified data group 111. Keywords are generated and classified every time data is added to the data store 11.

  Further, the data search device 1 includes a Japanese analysis engine 17 used for Japanese processing inside the data search device 1.

  Hereinafter, each functional block described above will be described in more detail.

○ Japanese analysis engine;
FIG. 5 is a block diagram showing a functional configuration of the Japanese analysis engine 17 that extracts and outputs a keyword from an input natural sentence.

  The Japanese analysis engine 17 includes a morphological analysis unit 171, an unnecessary word removal unit 172, and a duplicate word removal unit 173. The morpheme analyzing unit 171 divides the given natural sentence into words and outputs them to the unnecessary word removing unit 172 together with the part of speech information. The unnecessary word removal unit 172 refers to the part-of-speech information and the frequent dictionary 174, specifies an unnecessary word based on a predetermined rule, filters a given word, and extracts only an appropriate word as a keyword. The duplicate word removing unit 173 removes duplicate words included in the keyword extracted by the unnecessary word removing unit 172 and outputs the result.

  Such a Japanese analysis engine 17 is used at the time of keyword generation in the query processing engine 13, the additional data processing engine 14, and the data processing engine 16.

  Note that the Japanese analysis engine 17 is appropriately changed depending on the language used in the data to be processed by the data search device 1. That is, when a language other than Japanese is used in the data to be processed, the language analysis engine for the language is used instead of the Japanese analysis engine 17. Of course, it is not impeded that a language analysis engine corresponding to a plurality of languages is installed in the data search apparatus 1 instead of the Japanese analysis engine 17.

○ Data processing engine;
FIG. 6 is a block diagram showing a functional configuration of the data processing engine 16.

  The data processing engine 16 includes a keyword generation unit 161 and a keyword classification unit 162. The keyword generation unit 161 generates a keyword from each data in the data group 111 using the Japanese analysis engine 17 and outputs the keyword to the keyword classification unit 162. The number of keywords generated from one data is not limited to one and may be plural. Further, it is not prevented that duplicate keywords are generated from different data.

  The keyword classification unit 162 clusters keywords and determines classification items, and stores the keywords in the data store 11 according to the determined classification items. Clustering in the keyword classification unit 162 is performed based on the classified data group 111. In other words, the clustering in the keyword classification unit 162 is performed using the classification information of the data group 111. Although the clustering method is not limited, a method using a neural network, a genetic algorithm, link analysis, or the like is applicable. Hereinafter, an example in which a self-organizing map that is one form of a neural network is used for clustering among these methods and the like will be described.

  Depending on the clustering method, the keyword may need to be expressed numerically, or the processing may be easier if the keyword is expressed numerically. In such a case, the numerical value expressing the keyword is not limited to a single numerical value but may be a set of a plurality of numerical values. Also called). As a method of digitization, there are a method of digitizing based on the weight of the keyword in each data of the data group, a method of digitizing based on the co-occurrence degree of the keyword in each data of the data group 111, and the like. In the following, as an example of the former, the description will be made assuming that a method of digitizing a keyword based on the TF × IDF value of the keyword in each data of the data group is adopted. Here, “TF (Term Frequency)” is an appearance frequency (index word frequency) in keyword data, and “IDF (Inverse Document Frequency)” is an inverse of the number of data including the keyword. The TF × IDF value is a product of local weighting and global weighting, and serves as an index indicating the importance of keywords in each data.

When the keywords quantify (vectorization), first, TF × IDF value L (m, n) of the keyword K _n in the data D _m is calculated. The keyword K _n is expressed by an Nc-dimensional vector shown in Equation 1 using the calculated TF × IDF value L (m, n).

Notation in each component of the vector shown in Equation _{1 m∈G i (i = 1,2,} ···, Nc-1) , for the data D _m belonging to the classification item G _i, TF × IDF value L (m, n) means to add. In other words, each component of the vector shown in Equation 1 corresponds to the sum of the TF × IDF value L (m, n) for each classification item to which the data belongs. The set of vectors K _n obtained in this way is normalized. Specifically, normalization is performed so that the maximum value of the elements of the matrix shown in Equation 2 is 1 and the minimum value is 0.

  After this normalization, the magnitude of the component of the vector representing the keyword (hereinafter also referred to as “keyword vector”) is the importance of the keyword in the classification item corresponding to the component, in other words, the relevance of the keyword. It is about.

The normalized Nk vectors are input to the self-organizing map SOM illustrated in FIG. 7 and used for keyword categorization. The input layer IL of the self-organizing map SOM includes Nc units U11 to U14 (an example of Nc = 4 is shown in FIG. 7). Units U11 to U14 correspond to each component of the keyword vector Ki that is input data of the self-organizing map SOM. The output layer OL of the self-organizing map SOM is composed of Ng units U21 to U29 (in FIG. 7, an example of Ng = 9 is shown). The units U21 to U29 are arranged in a matrix having (Ng) ^1/2 units on one side. Each unit U21~U29 of the output layer OL is coupled by the propagation coefficient S _ij and each unit U11~U14 input layer IL. Here, the index i is an index that specifies a unit (component of the keyword vector K _n ) of the input layer IL, and the index j is an index that specifies a unit (classification item) of the output layer OL.

  In the above description, a keyword is added to the data store 11, but a key sentence composed of a plurality of words may be stored in the data store 11 instead of the keyword.

○ Query processing engine;
FIG. 8 is a block diagram showing a functional configuration of the query processing engine 13.

  The query processing engine 13 includes a query analysis unit 131. The query analysis unit 131 generates a keyword from the query using the Japanese analysis engine 17 and outputs the keyword to the keyword search engine 132 and the data search engine 15. The keyword generated by the query analysis unit 131 is an example of new data different from the data group 111 and the keyword group 112.

The keyword search engine 132 accesses the data store 11 and acquires the same classification keyword. For example, if the keyword classification is in the state shown in FIG. 4, the keyword search engine 132, the keyword K ₈ is given, the keyword K _20, ..., K _50, data store ... as the same classification keywords 11 from. In other words, the keyword search engine 132 specifies the classification item to which the given keyword belongs based on the classified keyword group 112 (using the classification information of the keyword group 112). That is, the data search apparatus 1 can effectively use the classification information of the data group 111 for the processing of the generated keyword (acquisition of the same classification keyword or generation of hints) via the classification information of the keyword group 112. It has become. Since keywords belonging to the same classification item have similar importance trends with respect to the data group 111, it can be said that there is a high possibility that an appropriate search result can be obtained by using the same classification keyword in a query.

  The hint generation unit 133 generates a hint that is auxiliary information related to the query based on the given same classification keyword, and outputs the hint to the user interface 12. Specifically, the hint generation unit 133 performs processing such as generating a hint that prompts the operator to input a query including a given keyword, and outputs the hint to the user interface 12. An example of such a hint is shown in an embodiment described later. Accordingly, the operator can input a query that leads to an appropriate search result using the same classification keyword to the data search device 1.

○ Additional data processing engine;
FIG. 9 is a block diagram showing a functional configuration of the additional data processing engine 14.

  The additional data processing engine 14 includes an additional data analysis unit 141 and an additional data classification unit 142.

  The additional data analysis unit 141 generates a keyword from the additional data using the Japanese analysis engine 17 and outputs the keyword to the additional data classification unit 142. The keyword generated by the additional data analysis unit 141 is an example of new data different from the data group 111 and the keyword group 112.

  The additional data classification unit 142 accesses the database, identifies the classification item to which the generated keyword belongs based on the classified keyword group 112 (using the classification information of the keyword group 112), and the additional data The classification item to which it belongs is determined as the identified classification item. The additional data for which the classification item is determined is stored in the data store 11 according to the classification item. That is, even if the additional data classification item is not specified, the data search device 1 can autonomously execute the additional data classification process using the keyword included in the additional data. In other words, the data search device 1 can be effectively used for the processing of the keyword that generated the classification information of the data group 111 (the classification of the additional data of the generation source).

<Operation>
○ Overall operation;
FIG. 10 is a flowchart showing an overall operation flow of the data search apparatus 1.

  In the first step S1 of the operation flow, either search or data addition is selected. If the search is selected, the operation flow moves to step S2 and the search operation subroutine is executed. On the other hand, when data addition is selected, the operation flow moves to step S3, and a subroutine for data addition operation is executed. After step S2 or S3 ends, the operation flow ends. In addition, the specific method of selection in step S1 is illustrated in the Example mentioned later.

○ Search operation;
FIG. 11 is a flowchart showing an operation flow of a subroutine for the search operation.

  In the first step S101 of the operation flow, query input is detected. When a query input is detected, the operation flow moves to the next step S102. When the query input is not detected, the operation flow returns to step S101, and the query input standby state continues. A specific method for inputting a query is exemplified in an embodiment described later.

  In step S <b> 102, the query analysis unit 131 generates a keyword from the query and outputs it to the keyword search engine 132 and the data search engine 15.

  Subsequently, the data search engine 15 performs a data search and outputs the search result to the user interface 12 (step S103).

  Subsequently, the keyword search engine 132 performs keyword search, extracts the same classification keyword (step S104), and the hint generation unit 133 generates a hint and outputs it to the user interface 12 (step S105).

  In step S106 following step S105, the search results and hints are presented to the operator using the user interface 12. After that, the operation flow of the search operation subroutine ends. In addition, the specific method of presentation of a search result and a hint is illustrated in the Example mentioned later.

  Through the above operation flow, the operator obtains a search result using the input query and obtains a hint related to the query.

○ Data addition operation;
FIG. 12 is a flowchart showing an operation flow of a subroutine of data addition operation.

  In the first step S201 of the operation flow, input of additional data is detected. When the input of additional data is detected, the operation flow moves to the next step S202. If the input of additional data is not detected, the operation flow returns to step S201, and the standby state for input of additional data continues. A specific method for inputting the additional data is exemplified in an embodiment described later.

  In step S <b> 202, the additional data analysis unit 141 generates a keyword from the additional data and outputs the keyword to the additional data classification unit 142.

  Subsequently, the additional data classification unit 142 performs data search to identify the classification item to which the given keyword belongs (step S203), and stores the additional data in the data store 11 based on the classification item identified by the additional data classification unit 142. (Step S204). After that, the operation flow moves to a step.

  Through steps S201 to S204, the data search apparatus 1 automatically classifies the additional data for which no classification item is designated and stores it in the data store 11.

  Step S205 is a data processing subroutine for keyword classification. After step S205 ends, the operation flow of the subroutine for data addition operation ends.

○ Data processing operation;
FIG. 13 is a flowchart showing an operation flow of a subroutine of data processing operation.

  In the first step S301 of the operation flow, the keyword generation unit 161 generates a keyword from each data in the data group 111.

Subsequently, the keyword classification unit 162 initializes the propagation coefficient S _ij by a predetermined method, and initializes the number of corrections s of the propagation coefficient S _ij (s = 0) (step S302).

Subsequently, the keyword classification unit 162 vectorizes all the keywords K _n (step S303) and sequentially inputs them to the self-organizing map SOM (step S304).

In step S305 following step S304, the keyword classification unit 162 corrects the propagation coefficient S _ij (step S305). Specifically, keyword classification unit 162, for all units U21~U29 output layer OL, the propagation coefficient S and the value unit U1~U4 holds the input layer IL (component value K _ni keyword vector K _n) The total error T with _ij (see Equation 3) is calculated, and the winner unit with the minimum total error T is specified.

Then, the keyword classification unit 162 corrects the propagation coefficient S _ij at a constant rate so that the total error T becomes small for the units in the vicinity region centered on the winner unit. Thereby, learning in the self-organizing map SOM has progressed.

  Subsequently, the correction count s is incremented (s ← s + 1) (step S306).

  In the subsequent step 307, a branching process is performed depending on whether the number of corrections s has reached the predetermined number N. If it has reached, the operation flow moves to step S309. On the other hand, if not reached, the operation flow proceeds to step S308.

  In step 308, the error correction ratio is reduced and the neighborhood area is reduced, and the operation flow returns to step S304. As a result, the learning of the self-organizing map SOM is continued while the progress of learning is sequentially suppressed until the number of corrections s reaches the predetermined number N.

In self-organizing map SOM step after learning completion of S309, the keyword classification unit 162 inputs a keyword vector k _n to the self-organizing map SOM. Further, the keyword classification unit 162 identifies a winner unit as in step S304 (step S311). Classification item corresponding to the winner unit is classified items belonging keyword k _n.

  In step S312, the keyword kn is stored in the data store 11 based on the classification item determined in step S311.

  With such an operation flow, the keyword group 112 extracted from the data group 111 is automatically classified based on the classification information of the data group 111.

  The first embodiment is an embodiment showing an application example of the data search apparatus 1 to a paper search system. Example 2 is an example showing an application example of the data search apparatus 1 to a medical support system. In the paper search system and the medical support system, the data group 111 and the keyword group 112 stored in the data store 11 are a specific data group and a keyword group to be processed by each system. In the paper search system and the medical support system, a user interface 12 suitable for the purpose of use of each system is provided. Hereinafter, the data group 111, the keyword group 112, and the user interface 12 will be described. In addition, Example 1 and Example 2 are only illustrations of the application method of the data search device 1. Therefore, the application method and application object of the data search device 1 are not limited by the first embodiment and the second embodiment, and can be arbitrarily changed within the scope of the claims.

[Example 1]
Example 1 relates to a paper search system. In the paper search system, the data group 111 is a set of text data (hereinafter, also referred to as “paper data”) relating to academic papers in the medical field (hereinafter, also referred to as “papers”). The paper search system searches the data group 111 using the input query, and presents paper data that matches the query to the operator as a search result. The paper search system stores the input additional paper data in the data store 11.

  Furthermore, the paper search system presents the operator with hints for changing a query to increase the possibility that desired paper data will be searched and output when executing the search. Further, the paper search system can determine the classification item to which the paper data belongs even if the classification item is not specified when the paper data is added. These support functions of the operator are realized by using the classification information of the data group 11 classified into a plurality of classification item groups.

  In the paper search system, the author name, target region, disease name, and session are adopted as classification criteria for classifying the data group 11. However, these classification criteria are examples, and changing these classification criteria to other classification criteria and adding other classification criteria to these classification criteria is not prevented.

  In the paper search system, a GUI (Graphical User Interface) using display on the display 47 is employed as the user interface 12. Therefore, the operator of the paper search system can give an instruction to the paper search system by performing a predetermined operation with the input device 45 while referring to the screen displayed on the display 47.

<Screen transition>
FIG. 14 is a screen transition diagram showing the transition of the screen displayed on the display 47 of the paper search system. Hereinafter, the operation of the paper search system will be described with reference to the screen transition diagram of FIG. 14 and the screen examples of FIGS.

○ Menu screen;
The menu screen SC11 illustrated in FIG. 15 is displayed when the paper search system is activated. Menu screen SC11 includes transition links 201 and 202 to search screen SC12 and registration screen SC13. When an operation for selecting one of the transition links 201 and 202 is performed on the menu screen SC11, the display on the display 47 is switched to a screen (search screen SC12 or registration screen SC13) corresponding to the selected transition link. Note that the selection of the transition link on the menu screen SC11 corresponds to the selection in step S1 of the flowchart of FIG.

○ Search screen;
The search screen SC12 illustrated in FIG. 16 is a screen for query input. Search screen SC12 includes a text area 211 for query input and a search button 212. The operator presses the search button 212 after inputting the query to the text area 211 to cause the paper search system to execute paper data search using the query. That is, in the paper search system, when pressing of the search button 212 is detected on the search screen SC12, paper data search and hint generation are performed according to the flowchart of FIG. After completion of the paper data search and hint generation processing, the display 47 is switched to the search result list display screen SC14. The pressing of the search button 212 corresponds to the query input in step S101 in the flowchart of FIG.

○ Search result list display screen;
A search result list display screen SC14 illustrated in FIG. 17 is a screen for displaying a list of search results. The search result list display screen SC14 includes a search result list table 221 and a re-search button 223. The search result list table 221 displays a list of titles and categories of articles that have been searched for and output. The titles 222a to 222g of each paper listed are also transition links to the detailed information display screen SC15 corresponding to each paper. Therefore, the operator can call the detailed information display screen SC15 of the desired paper on the display 47 by selecting the transition link corresponding to the desired paper. The re-search button 223 is a GUI instruction member for giving an instruction for calling the re-search screen SC16 to the paper search system.

○ Detailed display screen;
A detailed information display screen SC15 illustrated in FIG. 18 is a screen that displays detailed information of a paper corresponding to the transition link selected on the search result list display screen SC14. The detailed information display screen SC15 includes a detailed display table 231, a re-search button 232, and a full text display button 233. The detail display table 231 displays a title, category, author, keyword, weird record, and number of pages. The re-search button 232 and the full-text display button 233 are GUI instruction members for giving an instruction to call the re-search screen SC16 and the full-text display screen SC17 to the paper search system, respectively.

○ Full text display screen;
The full text display screen SC17 illustrated in FIG. 19 is a screen that displays the full text of the paper displayed on the detailed information display screen SC15. The whole sentence is a document described in plain text not including format information, text or image including format information, and may be PDF (registered trademark) or the like. The full text display screen SC17 includes a full text 241 and a back button 242 of the paper. The return button 242 is a GUI instruction member for giving a call instruction for the detailed information display screen SC15 to the paper search system.

○ Re-search screen;
The re-search screen SC16 illustrated in FIG. 20 is a screen for query input for re-search. The re-search screen SC16 includes a text area 251 and a search button 252 similar to the search screen SC12. Although the query 251a used in the previous search is displayed by default in the text area 251, the operator can edit the query 251a by a predetermined operation.

  Furthermore, query change hints HT1 to HT5 are displayed below the text area 251 of the re-search screen SC16. The hints HT1 to HT5 are hints that prompt the user to change keywords (“Aoki”, “colon”, “cancer”, “recurrence rate”, and “colon cancer”) included in the query 251a used in the previous search. .

  The hint HT1 is a hint generated from the same classification keyword of “Aoki” (“Tomomi Aoki” and “Yutoshi Aoki”) related to the classification item of the author's name. The hint HT2 is a hint generated from the same classification keyword (“colon” and “rectum”) of “large intestine” related to the classification item of the target region. The hint HT3 is a hint generated from the same classification keyword (“malignant tumor”, “tumor marker”, and “radiation”) of “cancer” related to the disease name classification item. The hint HT4 is a hint generated from the same classification keyword (“survival rate” and “life expectancy”) of “recurrence rate” related to the session classification item. The hint HT5 is a hint generated from the same classification keyword (“colon cancer” and “colon cancer”) of “colon cancer” related to the session classification item. Since the keywords included in the hints HT1 to HT5 are similar in appearance tendency in the data group 111 to the keywords included in the query used in the previous search (hereinafter, also referred to as “previous keywords”), they are substituted for the previous keywords. It can be used as a keyword. Therefore, the operator may obtain a more appropriate search output by referring to the hint and changing the query.

  Further, the re-search screen SC16 includes an end button 253. The end button 253 is a GUI instruction member for instructing the paper search system to shift to the menu screen SC11. By pressing the end button 253, the operator can switch the display 47 to the menu screen SC11 again.

○ Registration screen;
A registration screen SC13 illustrated in FIG. 21 is a screen for paper registration. The registration screen SC13 includes a text box 261, a reference button 262, and a registration button 263. The operator inputs the file location (path) of the registered paper in the text box 261 (or presses the registration button 263 after pressing the file reference button 262 to select the registered paper), thereby pressing the registration button 263. Registration of data in the data store 11 is executed by the paper search system. That is, when pressing of the registration button 263 is detected on the registration screen SC13, the paper data of the path input in the text box 261 is stored in the data store 11 as additional data according to the flowchart shown in FIG. Note that FIG. 21 does not include a GUI operation member for designating classification items of paper data, but the paper search system can automatically classify input paper data. The pressing of the registration button 263 corresponds to the additional data input in step S201 in the flowchart of FIG.

  When the registration of the paper is completed after the registration button 263 is pressed, the display on the display 47 is switched to the registration success screen SC18.

○ Registration success screen;
A registration success screen SC18 illustrated in FIG. 22 is a screen for notifying the operator of the success of the paper registration.

  The registration success screen SC18 includes a character string 271 for notifying the operator of successful paper registration, and transition links 272 and 273 to the registration screen SC13 and the menu screen SC11. When an operation for selecting one of the transition links 272 and 273 is performed on the registration success screen SC18, the display 47 shifts to a screen corresponding to the selected transition link.

[Example 2]
Example 2 relates to a medical support system. In the medical support system, the data group 111 is a set of text data (hereinafter also referred to as “incident data”) related to an incident experienced by a medical worker when performing a medical practice. The medical support system performs a search on the data group 111 using the input query, and presents incident data that matches the query to the operator as a search result. In addition, the medical support system stores the input additional incident data in the data store 11. The query in the medical support system is generated based on the condition information input by the medical worker using the template on the condition input screen SC21.

  Furthermore, the medical support system presents a keyword that is closely related to the keyword included in the query to the operator as a hint when executing the search. In addition, the medical support system can determine the classification item to which the incident data belongs even if the classification item is not specified when incident data is added. These support functions of the operator are realized by using the classification information of the data group 111 classified into a plurality of classification item groups.

  Classification criteria for classifying the data group 111 include the department (internal medicine, surgery, otolaryngology, etc.) and the years of experience (generally attributes) of the medical staff who experienced the incident, the time of occurrence of the incident, The weather at the time of the incident is adopted. However, these classification criteria are examples, and changing these classification criteria to other classification criteria and adding other classification criteria to these classification criteria is not prevented.

  In the medical support system, a GUI using display on the display 47 is adopted as the user interface 12 as in the paper search system. Therefore, the operator of the medical support system can give an instruction to the medical support system by performing a predetermined operation with the input device 45 while referring to the screen displayed on the display 47.

<Screen transition>
FIG. 23 is a screen transition diagram showing transition of screens displayed on the display 47 of the medical support system. Hereinafter, the operation of the medical support system will be described with reference to the screen transition diagram of FIG. 23 and the screen examples of FIGS.

○ Menu screen;
A menu screen SC22 illustrated in FIG. 24 is displayed when the medical support system is activated. Menu screen SC22 includes transition links 301 and 302 to condition input screen SC21 and incident registration screen SC23. When an operation for selecting one of the transition links 301 and 302 is performed on the menu screen SC22, the display on the display 47 is switched to a screen (condition input screen SC21 or incident registration screen SC23) corresponding to the selected transition link. Note that the selection of the transition link on the menu screen SC22 corresponds to the selection in step S1 of the flowchart of FIG.

○ Condition input screen;
A condition input screen SC21 illustrated in FIG. 25 is a screen for inputting a condition of a medical worker.

  The condition input screen SC21 is a template having a plurality of input item frameworks. The template includes the department and years of experience of the medical staff who experienced the incident, the time when the incident occurred, the weather at the time of the incident, and the medical action related to the incident (in FIG. 25, the nurse performs the treatment on the patient). Text areas 311 to 315 and a search button 316 are input. However, the input item illustrated in FIG. 25 is an example, and it is not hindered to add another input item to the input item or change the input item to another input item. The operator presses the search button 316 after inputting to the text areas 311 to 315, thereby causing the medical support system to search for incident data corresponding to the input content. That is, when pressing of the search button 316 is detected on the condition input screen SC21, incident data search and hint generation are performed according to the flowchart of FIG. In the medical support system, a keyword is generated from a plurality of character strings input using a template. In addition to Japanese analysis, keyword information is generated using information on input items in the template. After the incident data search and hint generation, the display 47 is switched to the search result list display screen SC24. Note that pressing of the search button 316 corresponds to the query input in step S101 in the flowchart of FIG.

○ Search result list display screen;
The search result list display screen SC24 illustrated in FIG. 26 is a screen for displaying a list of search results. The search result list display screen SC24 includes a search result list 321 and a related keyword list 322. The related keyword list 322 displays a list of related keywords that are hints related to the query. The hint is the same classification keyword extracted by the keyword search engine 132. That is, the hint generation unit 133 of the medical support system outputs the same classification keyword to the user interface 12 without performing special conversion.

  The search result list 321 displays a list of care contents of the incident data searched for and output. The search output is a search output obtained by performing a search using a keyword directly generated from the input of the condition input screen SC21 and the same classification keyword of the keyword. Thus, by using the same classification keyword (related keyword) for the search, it is possible to prevent important incident data from falling out of the search output. The care content of each incident data listed in the search result list is also a transition link to the detailed information display screen SC25 corresponding to each incident data. Therefore, the operator can call the detailed information display screen SC25 of the desired incident data on the display 47 by selecting the transition link corresponding to the desired incident data.

  The search result list display screen SC24 includes a return button 323. The return button 323 is a GUI instruction member for giving a call instruction of the condition input screen SC21 to the medical support search system.

○ Detailed display screen;
The detailed information display screen SC25 illustrated in FIG. 27 is a screen that displays detailed information of incident data corresponding to the transition link selected on the search result list display screen SC24. Detailed information display screen SC25 includes a detailed display table 331 and a return button 332. The detailed display table 331 displays the department and years of experience of the medical staff who experienced the incident, the time when the incident occurred, the weather at the time of the incident, the contents of the medical practice related to the incident, and the contents of the incident Is done. The return button 332 is a GUI instruction member for giving a call instruction for the search result list display screen SC24 to the medical support system.

○ Incident registration screen;
The incident registration screen SC23 illustrated in FIG. 28 is an incident registration screen. The incident registration screen SC23 is a template, similar to the condition input screen SC21. The template includes text areas 341 to 346 for inputting the same items as those on the detailed display screen SC25. The operator causes the medical support system to add incident data to the data store 11 by pressing the registration button 347 after inputting the text areas 341 to 346. That is, when pressing of the registration button 347 is detected on the incident registration screen SC23, incident data input to the template is stored in the data store 11 as additional data. Note that FIG. 28 does not include a GUI operation member for specifying a classification item of incident data, but the medical support system can automatically classify input incident data.

  When the registration of incident data is completed after the registration button is pressed, the display on the display 47 shifts to a registration success screen SC26. The pressing of the registration button 347 corresponds to the additional data input in step S201 in the flowchart of FIG.

○ Registration success screen;
A registration success screen SC26 illustrated in FIG. 29 is a screen for notifying the operator of successful incident registration. The registration success screen SC26 includes a character string 351 for notifying the operator of the success of incident data registration, and a return button 352. When the return button 352 is pressed on the registration success screen SC26, the display 47 shifts to the incident data registration screen SC23.

FIG. 2 is a block diagram showing a hardware configuration of a computer 4 for realizing the data search device 1. 2 is a block diagram showing an overall functional configuration of the data search device 1. FIG. Data group 111 is a diagram showing _{{D 0, D 1, ···} , D p} is category G _0, G _1, ···, an example of a state of being classified in G _Nc-1. Keyword group 112 is a diagram showing _{{K 0, K 1, ···} , K Nk} classification item G _0, G _1, ···, an example of a state of being classified in G _Nc-1. 3 is a block diagram showing a functional configuration of a Japanese analysis engine 17. FIG. 2 is a block diagram showing a functional configuration of a data processing engine 16. FIG. It is a figure which illustrates self-organizing map SOM. 3 is a block diagram showing a functional configuration of a query processing engine 13. FIG. 3 is a block diagram showing a functional configuration of an additional data processing engine 14. FIG. 4 is a flowchart showing an overall operation flow of the data search device 1. It is a flowchart which shows the operation | movement flow of the subroutine of search operation | movement. It is a flowchart which shows the operation | movement flow of the subroutine of data addition operation | movement. It is a flowchart which shows the operation | movement flow of the subroutine of a data processing operation | movement. It is a screen transition diagram which shows the transition of the screen displayed on the display 47 of a paper search system. It is a figure which illustrates menu screen SC11. It is a figure which illustrates search screen SC12. It is a figure which illustrates search result list display screen SC14. It is a figure which illustrates detailed information display screen SC15. It is a figure which illustrates full-text display screen SC17. It is a figure which illustrates re-search screen SC16. It is a figure which illustrates registration screen SC13. It is a figure which illustrates registration success screen SC18. It is a screen transition diagram which shows the transition of the screen displayed on the display 47 of a medical assistance system. It is a figure which illustrates menu screen SC22. It is a figure which illustrates condition input screen SC21. It is a figure which illustrates search result list display screen SC24. It is a figure which illustrates detailed information display screen SC25. It is a figure which illustrates incident registration screen SC23. It is a figure which illustrates registration success screen SC26.

Explanation of symbols

IL input layer OL output layer U11 to U14, U21 to U29 Unit SOM Self-organizing map HT1 to HT5 Hint

Claims (22)

A data processing device,
Generating means for generating a second data group from a first data group classified into a classification item group including a plurality of classification items;
Classification means for classifying the second data group into the classification item group based on the classified first data group;
Determining means for determining a classification item in the classification item group to which new data not included in the first data group and the second data group belongs, based on the classified second data group;
A data processing apparatus comprising: The data processing apparatus according to claim 1,
Each data of the first data group is text data,
A data processing apparatus, wherein each data of the second data group is text data corresponding to an index of text data of the first data group. The data processing apparatus according to claim 2, wherein
The data processing apparatus, wherein the first data group is a data group related to a document, and the second data group is a data group related to an index of the document. The data processing apparatus according to claim 2, wherein
The data processing apparatus, wherein the first data group is a data group related to an incident caused by a medical practice, and the second data group is a data group related to an index of the incident. The data processing device according to any one of claims 1 to 4,
The data processing apparatus, wherein the second data group is partial data of the first data group. The data processing apparatus according to claim 2, wherein
The generating means
Morphological analysis means for dividing a character string described in the text data into word groups;
Construction means for constructing the second data group from the word group;
A data processing apparatus comprising: The data processing apparatus according to claim 6, wherein
The construction means is
A data processing apparatus comprising means for removing duplicate words in the word group. The data processing device according to claim 6 or 7,
The construction means is
A data processing apparatus comprising: means for specifying an unnecessary word in the word group according to a predetermined unnecessary word specifying rule and removing the unnecessary word. The data processing device according to any one of claims 6 to 8,
The classification means is
Calculating means for calculating a weight of each data of the second data group in each data of the first data group according to a predetermined weight calculation rule;
A data processing apparatus using the weight for classification of the second data group. The data processing apparatus according to claim 9, wherein
The data processing apparatus characterized in that the weight is the importance of data determined by a predetermined importance determination rule. The data processing apparatus according to claim 10, wherein
The degree of importance is a product of an appearance frequency of an index word and a reciprocal of the number of documents including the index word. The data processing apparatus according to any one of claims 1 to 11,
The classification means is
Vectorization means for converting each data of the second data group into a vector having a component corresponding to a classification item of the first data group;
A data processing apparatus using the vector for classification of the second data group. The data processing apparatus according to claim 12, wherein
A data processing device, wherein each component of the vector is determined for each data of the second data group based on a weight expressing a degree of relevance to each classification item of the first data group. The data processing device according to any one of claims 1 to 13,
A data processing apparatus, wherein the classification means uses a neural network for classification. The data processing apparatus according to claim 14, wherein
A data processing apparatus, wherein the neural network is a self-organizing map. The data processing device according to any one of claims 1 to 15,
The data processing apparatus, wherein the new data is data added to the first data group. The data processing device according to any one of claims 1 to 15,
The data processing apparatus, wherein the new data is generated from a query used for a search for the first data group. The data processing apparatus according to claim 17,
A data processing apparatus, further comprising an extracting unit that extracts data of another second data group belonging to the classification item to which the new data belongs. The data processing apparatus according to claim 18, wherein
An output means for outputting auxiliary information related to the new data;
The data processing apparatus, wherein the auxiliary information is created based on data of another second data group included in a classification item to which the new data belongs. The data processing apparatus according to claim 1,
The data processing apparatus further comprising input means for allowing an operator to input the new data. A data processing method,
Generating a second data group from the first data group classified into a classification item group including a plurality of classification items;
A classification step of classifying the second data group into the classification item group based on the classified first data group;
A determination step of determining a classification item in the classification item group to which new data not included in the first data group and the second data group belongs, based on the classified second data group;
A data processing method comprising: A data processing program, the computer executing the program,
A generating step of generating a second data group from the first data group classified into a classification item group including a plurality of classification items;
A classification step of classifying the second data group into the classification item group based on the classified first data group;
A determination step of determining a classification item in the classification item group to which new data not included in the first data group and the second data group belongs, based on the classified second data group;
A data processing program characterized by causing

Images